open/portfolio-optimiser-claude
1
0
Fork 0
Code Issues Releases 1 Activity
1 release 1 tag
  • v0.1.0 0b69354c74

    v0.1.0 - the portfolio-optimiser method on the Claude Agent SDK Stable

    ktg released this 2026-08-17 11:33:27 +00:00 | 6 commits to main since this release

    No known key found for this signature in database

    The portfolio-optimiser method, implemented on the Claude Agent SDK (sibling implementation D7), built from the shared frozen spec and its golden suite.

    What is in it

    • Deterministic backbone: the typed cost-IR, the mandatory blocking validator (frozen by the shared golden suite, its only oracle), first-class provenance, fail-fast startup contracts.
    • Agentic loop: bounded generation, maker-checker debate, the validator gate, informed refinement; the budget meter admits no unbounded loop, with an optional pre-call run-total USD belt.
    • Learning loop: the OKF context seam (navigation, never chunk-stuffing), the ExpeL-style experience fold, the async expert-verdict inbox, the fail-closed promotion gate.
    • Ingest layer: deterministic CSV (file) and SQLite (sql) connectors, frozen by byte-identical golden extractions. http/MCP is a named, unbuilt extension point and is rejected fail-fast.
    • Value layer: the fail-closed savings ledger, hard/soft goal contract, outbox, HITL view, pre-run cost simulation, SDK/API preflight, opt-in notification sinks, per-run value report.
    • Operator CLI: one collecting entrance (run.py) for a single project or a portfolio, plus standalone entrances for valuereport, hitl, costsim, preflight.
    • Load-bearing tests: every seam is proven by a test that goes red when the seam is detached. The whole suite (795 tests) runs offline, with no API key and no network.

    Why 0.1.0 and not 1.0.0

    Two gaps are documented and deliberately held open rather than papered over. The shared spec's V1 generated shape has landed upstream but is not adopted in the goldens here: the golden is the library's own emission, so adoption is gated on the ingest-library pin swap, and a ratchet test goes red the moment a materialize() run reaches the new shape. And the SDK pin (>=0.2.111,<0.3) reaches further than the range whose premises are source-verified (through 0.2.110). A 1.0.0 would claim a settled surface this implementation does not yet have.

    See CHANGELOG.md for the full entry.

    Downloads